Machine learning for fast motion compensated quantitative abdominal DCE-MRI

Functional imaging with dynamic contrast-enhanced MRI (DCE-MRI) provides physiological markers of permeability, perfusion and glomerular filtration rate (GFR), a measure of kidney function. One of the most important applications of DCE-MRI is to assess GFR in pediatric hydronephrosis patients with obstructed kidneys. In the absence of GFR information, children who stand to benefit from immediate surgical reconstruction might be overlooked or delayed in receiving treatment, and those who might benefit from a more conservative approach (i.e., “watchful waiting”) might receive an unnecessary surgical intervention. The current standard, nuclear renography (MAG3) is slow and low resolution, and delivers potentially harmful ionizing radiation.

Current methods of DCE-MRI in neonates and children are less than optimal, and therefore, DCE-MRI is underutilized in clinical practice. Technical challenges include insufficient temporal resolution to capture fast arterial input function (AIF) dynamics, unavoidable respiratory motion and bulk motion, and a lack of automated post processing techniques for accurate computation of markers.

The primary objectives of this project are three-fold. First, to develop and evaluate a new bulk and respiratory motion-compensated, high spatiotemporal resolution DCE-MRI technique for accurate estimation of functional markers. Second, to further improve the robustness and speed of DCE-MRI using a fast, deep learning (DL) technique with integrated temporal prior for the reconstruction of motion-compensated, higher quality, high temporal resolution images. Third, to develop an automatic quantitative analysis pipeline including segmentation and tracer kinetic model-fitting using DL techniques for fast, robust and accurate quantification of functional markers.

The successful completion of these aims will provide new, clinically important abdominal imaging capabilities, with real-time, motion-compensated image reconstruction and reliable real-time parameter estimation from high temporal and spatial resolution DCE-MRI. This work will extend the usefulness of DCE-MRI to pediatric patients who are unable to remain still in the scanner, and eliminate the need for repeated scans and sedation in infants.